AI use drives home pages below accessibility standards

Websites are increasingly failing accessibility standards as developers and content teams use AI to generate copy and code. New figures from WebAIM suggest the problem now affects most home pages.

WebAIM found that more than 95% of sites in its latest study did not conform to Web Content Accessibility Guidelines 2.2. The findings point to persistent problems with basic web elements, including image descriptions, link text and colour contrast.

Accessibility has both usability and legal implications in Australia. Government and government-funded websites must meet WCAG standards, while broader obligations also arise under the Disability Discrimination Act 1992.

Data from the Australian Bureau of Statistics shows that 21.4% of Australians live with disability. Failures in website accessibility can therefore affect a large share of the population trying to access information, services and transactions online.

Dan Kaufman of web writing training business Media Survival said the growing use of AI in publishing workflows was contributing to errors that require human judgement to avoid.

"Making sites accessible requires understanding both context and what users need, and AI struggles with this," said Dan Kaufman of Media Survival.

He cited alternative text for images as one of the clearest examples. Alt text is meant to convey the information an image provides to users of screen readers or braille readers, rather than simply describe what it looks like.

"The web was designed to be accessible through HTML. For example, if you have an image on your site, you can add alternative text to its code for people with visual impairments.

"Yet writing this alternative text is harder than most people realise, because it is not meant to just describe how the image looks. Instead, alt text should give users the key information the image is conveying. Describing a chart as blue and pie-shaped is different from explaining the information in it.

"Knowing why an image is being used requires human judgement, and AI does not possess that. Images that are purely decorative and not informative are meant to be ignored by assistive technology such as screen readers or braille readers. Yet AI does not know what is or is not informative, so it will create slop for the alternative text instead," Kaufman said.

Link problems

The WebAIM study also highlighted weaker performance in link text. It found that 15.2% of pages had ambiguous link labels such as "click here" and "more", up from 13.7% the previous year.

Such wording can create difficulties for users of assistive technology, which often reads links without surrounding page context. Clear labels help users understand where a link goes before selecting it.

"It's ridiculous that, in 2026, many large websites still have 'click here' links," Kaufman said.

He said the spread of AI tools across content and publishing teams made it more important for staff to understand the basics of accessibility, search and web usability rather than rely on automated output.

"Anyone who writes for a site, or uploads content such as images and video, needs to understand the fundamentals of web usability, SEO and accessibility.

"AI automates many tasks, such as writing alt tags and creating meta descriptions for SEO, and it often gets them wrong. I'm not saying people shouldn't use AI, but at the very least they need the knowledge to assess what it has done," Kaufman said.

Contrast issues

Low-contrast text remained the most common accessibility problem in the WebAIM data. This happens when text does not stand out enough from the background or image behind it, making reading difficult for many users.

The study found low-contrast text on 83.9% of home pages, up from 79.1% a year earlier. The issue is common on visually led pages where design choices override legibility.

"I see this mistake all the time," Kaufman said.

"Websites aren't just meant to be pretty. They're meant to be effective, and that means they need to be readable by everyone, since not everyone has perfect vision. And even if they did, no one wants to squint at a screen to try to read the text," he said.

SEO overlap

The same pattern is appearing in search-related metadata, according to Kaufman. Website teams are using AI to draft title tags and meta descriptions, but those results often fall short when left unreviewed.

That overlap matters because both accessibility and search depend on accurate language, clear structure and an understanding of what users are trying to do. Automated text may fill a field, but it does not necessarily meet that standard.

"Although AI-generated meta descriptions and title tags are usually better than AI-generated alt tags, they're still not ideal," Kaufman said.

"In my experience, they work as a placeholder at best until you can do a proper job. That's because SEO isn't about stuffing keywords onto a page: as with accessibility, it's about understanding what people are searching for, what they need, and how your page meets that. And that still requires a human, not a bot, to apply critical thought," he said.